Open guoquan opened 4 years ago
What should be the semantic of 'index'
now? For anything generated by the candidate sensor, it is just tensor(1)
, which is a result of True
. But basically it will always be True
(in a datanode) because the datanodebuilder recognizes True
and build the datanode.
It will be more meaningful if it is the "index" among the adjacency matrix for the connecting relation.
For example, a relation's index
indicating the head span and tail span, a token's index
indicating the sentence id and token 'index' in the sentence.
(Proposal)
Here is a proposal for index semantic. This is slightly different than today's discussion.
Since we have decided to use them as identifiers, and we have a clear mapping from the tensor to datanode based on #213 , we can use the index in the list/vector/matrix as the index for each datanode. So it is clear whenever a list/vector/matrix is given, we need to retrieve the index
th row and that is corresponding to the datanode.
That means, we cannot allow users to customize this index. @auszok can generate it as interpreting the first sensor assigning value to the concept and initiate the index.
Discussion about if we should have any default property like 'index' or 'id'? When should we need this default? When should the user be able to not specify it?